期刊名称：Bulletin of the Polytechnic Institute of Jassy, Constructions, Architechture Section
印刷版ISSN：1224-3884
电子版ISSN：2068-4762
出版年度：2015
卷号：LXV
期号：3
页码：83-92
出版社：Gheorghe Asachi Technical University of Jassy
摘要：This paper proposes a manner to verify the concrete samples homogeneity using artificial neural networks. This method determines the percentages of different areas of component materials visible at top and bottom of a concrete cylinder having 20 cm diameter and 20 cm height. The materials that have been achieved are rubber grains, aggregates and mineral matrix. The training of the neural network was realise by using backpropagation algorithm and then, in order to separate the regions of interest was used Levenberg – Marquardt algorithm. As a neural network input data were used photos having 258 ×170 pixels resolution achieved both on the both sides of the cylinder.
